Post BA9TUOo0Arv84GofrM by roytam1@mastodon.ktachibana.party
(DIR) More posts by roytam1@mastodon.ktachibana.party
(DIR) Post #BA9BYtwmXB2qdRv7WS by niconiconi@mk.absturztau.be
1 likes, 1 repeats
Hacker News comments of "Why TCP over TCP is a bad idea (2001)" https://news.ycombinator.com/item?id=25080693The ignorance of networking is staggering here. I fully support Trump to build the Great Firewall of America to teach developers networking common sense. One user has never heard about TCP-in-TCP penalty at all, and called this fact "fascinating". Another user doesn't understand the different between TCP-in-TCP and SSH port forwarding (via -L, -R, or -D), the latter is a Layer 7 application proxy just like Nginx, HAProxy, or a HTTP/SOCKS proxy in your Web browser (a "shim layer" between 4 and 7 if you want to be pedantic like RFC 1928, since it doesn't parse the content), there no TCP-in-TCP penalty unless you do TUN/TAP. Yet another user said "it's strange" when OpenVPN over UDP performed poorly, but the performance is great over TCP. It's common knowledge that "unclassified UDP" is deprioritized to death by many ISPs because it's often used for latency-tolerant bulk file transfers or brute-force flows that don't implement congestion control at all.
(DIR) Post #BA9CNR8pOPFq44eTlg by PeterCxy@comfy.social
0 likes, 1 repeats
@niconiconi@mk.absturztau.be I keep finding it difficult to explain to people why L4 proxying TCP is okay but tunneling TCP over L3 or L2 VPN on TCP is not
(DIR) Post #BA9I49wZhyd8wy6zLc by niconiconi@mk.absturztau.be
0 likes, 0 repeats
@PeterCxy@comfy.social It also makes you wonder: people say tunneling TCP is "not okay" because it stacks TCP on TCP, while people say proxying TCP is "okay" because there's only one connection between endpoints like "normal" TCP. But there's still a problem, a proxy causes "double-buffering". So what does Layer 7 bufferbloat do to the proxy's connection quality (interactiveness) when there's bandwidth mismatch? HTTP/2 has streams, frames, priority, a dependency tree, and a scheduler, so the interaction can be pretty complicated. Bufferbloat problems are studied nearly exclusively on L3/L4, "L7 bufferbloat for proxied and multiplexed connection" would be a "fundamental research" question for people with too much time.
(DIR) Post #BA9TUOo0Arv84GofrM by roytam1@mastodon.ktachibana.party
0 likes, 0 repeats
@niconiconi @PeterCxy Had fun on dynamic tunneling on SSH on stunnel over corp's HTTPS proxy to phone home.